Three artificial intelligence data challenges based on CT and MRI

Area of Science:

• Medical imaging analysis
• Artificial intelligence in radiology
• Public health research

Background:

• The French Society of Radiology organized the second edition of an artificial intelligence (AI) data challenge.
• The challenge aimed to address public health issues using AI and high-quality, multicenter medical imaging databases.

Purpose of the Study:

• To develop AI tools for relevant public health issues.
• To create large, multicenter, high-quality medical imaging databases.
• To incorporate three-dimensional (3D) imaging information and prognostic questions into AI development.

Main Methods:

• Clinical questions were proposed by radiology subspecialty colleges and assessed for AI feasibility.
• A secure platform was established for anonymized data upload, adhering to data protection regulations.
• Radiologists performed weekly quality checks and annotations on the anonymized examinations.

Main Results:

• Three distinct AI challenges were conducted, focusing on pulmonary nodule detection (3D CT), multiple sclerosis disability prediction (3D MRI), and sarcopenia estimation (2D CT).
• A total of 4347 examinations were collected from 24 centers, forming three independent databases with a 94% inclusion rate.
• 143 participants in 20 multidisciplinary teams competed in the challenge.

Conclusions:

• Three successful AI data challenges were completed, each involving over 1200 general data protection regulation-compliant CT or MRI examinations.
• Future challenges should incorporate more complex scenarios, integrating histopathological or genetic data to better reflect clinical practice.